quiz bu lëkkale ak tegtal · Diggante Niveau

Lens Logit ak Quiz Lens Tuned

Saytu liñu jàng ci diggu couche bi, structure residuel bi, tekkikatu lens biñ akorde ak ay gàttal ci lens logit bu ñor bi.

Yooni njiit yu jëm ciLentille Logit Ak Lentille Tuned
Laaj 1 de 8

Lu lens logit di def?

Xoolaatal laaj bu nekk8 laaj

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

  1. Lu lens logit di def?
    • Jëfandikool model bi mujjee génne ci diggu réew yu nëbbu yi ngir jàng xam-xam bu nekk ci couche bu nekk
    • Yokk ay diisaay yu bees ci transformatër bi
    • Dindi bàyyi xel ci model bi
    • Tekki làkki nit ñi
  2. Ban yokkute bu am solo la lens biñ akorde yokk ci lens logit bu ñor bi?
    • Dafay jëfandikoo vocabulaire bu gëna mag
    • Dafay tàggat ab kàrt affine bu ndaw ci layer bu nekk ngir tekki réew yu nëbbu yi ci base bu mujj bi
    • Dafay dindi bépp njariñu dindi
    • Ci modelu nataal rek lay dox
  3. Lan mooy tax lens yooyu mëna am ci transformateur yi?
    • Seggal yiy boole
    • Tokenisatër bi
    • Flux residuel biy yokk bi ñuy génne ci logits yi
    • Tolluwaayu jàng bu optimisatër bi
  4. Kan moo njëkka dugal lens logit bi, ak kañ la?
    • xel mu xóot ci 2015
    • 2018
    • 2012
    • 2020
  5. Fan la lens logit bi di faral di wane tontu bu mujj bi 'crystallizing'?
    • Ci diggu-ba-mujj
    • Ci layer bu njëkk bi kese
    • Musu, tontu bi dafay feeñ benn yoon
    • ci biti model bi lëmm
  6. Lan moo waral lens logit bu ñor bi mënul wóolu ci diisaay yu njëkk yi?
    • Couche yu njëkk yi amul ay neuron
    • Seen nekkin yu nëbbu yi ba tay méngoo wuñu ak li ñuy seentu ci génne gu mujj gi
    • Layers yu njëkk yi duñu musa jëfandikoo ci wax luy am
    • matrix bi amul benn njuumte la
  7. What does the model’s final unembedding map a residual-stream vector into?
    • A new set of transformer layers
    • A similarity graph over training examples
    • Vocabulary logits that can be converted into token probabilities
    • A gradient clipping threshold
  8. Why can an early-layer raw logit-lens prediction be misleading?
    • The layer has already generated the full answer
    • The tokenizer has been removed
    • The unembedding only works on images
    • The intermediate state may not yet be aligned with the final output basis